A multi-dimensional parameter perception-based bulk and general cargo terminal operation whole-process management and control system

Abstract

This invention relates to the field of port automation technology, specifically to a multi-dimensional parameter perception-based end-to-end control system for bulk cargo terminal operations. It includes: a multi-source parameter acquisition unit; a multi-dimensional parameter adaptive fusion unit; a process node dynamic analysis unit; an intelligent decision-making and instruction generation unit; and a closed-loop control and verification unit. This invention comprehensively collects multi-dimensional data through the multi-source parameter acquisition unit; simultaneously, based on an adaptive confidence model improved by Kalman filtering, it configures a constraint confidence time window for various parameters, combines time decay confidence weights and operational reliability factors to complete time dimension calibration, and further supplements this with physical constraint noise reduction and electromagnetic environment calibration to achieve accurate fusion of multi-source heterogeneous data. Finally, the process node dynamic analysis unit completes accurate adaptation to multiple scenarios by pre-stored standard process diagrams for various operations such as ship loading / unloading, train loading / unloading, and vehicle port access, thus overcoming the deficiency of existing technologies that do not cover all business scenarios.

Description

Technical Field

[0001] This invention relates to the field of port automation technology, and more specifically, to a full-process control system for bulk cargo terminal operations based on multi-dimensional parameter perception. Background Technology

[0002] The current production management system for bulk cargo terminals suffers from pain points such as inconsistent data standards, prominent data silos, frequent manual intervention, and lack of closed-loop control, making it difficult to meet the needs of refined management and efficient operation.

[0003] In the existing technology, relevant patents have already explored the field of terminal operation management. For example, Chinese patent CN202410124631.8 discloses an automated container terminal operation management system, which includes automated equipment, manned trucks, yard area management, and TOS system units, enabling real-time monitoring of equipment operation and yard congestion, thereby improving the efficiency and accuracy of central control management. Another example is Chinese patent CN202511452429.9, which discloses a dry bulk cargo loading terminal operation scheme. By using yard gridding, laser point cloud scanning to update status, and combining simulation models to dynamically adjust operation strategies, it solves the problem of slow response in traditional static management.

[0004] Despite the design advantages of the aforementioned technical solutions, they also suffer from the following technical shortcomings: First, limited scenario adaptability: Chinese patent CN202410124631.8 focuses on equipment and truck management in automated container terminals, failing to cover core business modules specific to terminals, such as business documentation, commercial billing, and piece-rate wages; Chinese patent CN202511452429.9 addresses the optimization of material handling and scheduling at dry bulk cargo loading terminals, but cannot adapt to complex scenarios involving multiple product categories and operational processes, thus failing to achieve integrated management of all terminal operations; Second, insufficient data governance and closed-loop management: Chinese patents CN202410124631.8 and CN202511452... 429.9 lacks a unified data naming and definition standard, making it difficult to ensure data consistency across production, finance, and scheduling stages. It also fails to form a closed-loop data mechanism encompassing "collection-processing-application-feedback," thus failing to meet the demands of refined terminal management for controllable data quality and traceable accountability. Thirdly, it lacks end-to-end intelligence and multi-terminal collaboration: Chinese patents CN202410124631.8 and CN202511452429.9 do not address core intelligent requirements such as group-level rate standardization, automatic billing, and intelligent scheduling algorithm models. They also fail to consider multi-terminal collaborative applications involving mobile and mechanical terminals, resulting in a high degree of manual intervention and hindering the development goal of efficient terminal operations. Therefore, we propose a multi-dimensional parameter perception-based end-to-end control system for bulk cargo terminal operations. Summary of the Invention

[0005] The purpose of this invention is to provide a full-process control system for bulk cargo terminal operations based on multi-dimensional parameter perception, in order to solve the problems mentioned in the background art, such as limited scenario adaptability, insufficient data governance and closed-loop control, and lack of full-chain intelligence and multi-terminal collaboration.

[0006] To address the aforementioned technical problems, the present invention aims to provide a multi-dimensional parameter sensing-based end-to-end control system for bulk cargo terminal operations, comprising:

[0007] The multi-source parameter acquisition unit adopts a multi-source architecture that combines industrial bus, distributed sensing, machine vision, UWB positioning and NFC technology to collect equipment operating parameters, environmental and weighing data, cargo information, personnel and vehicle trajectories, cargo handling and shift data, and output multi-source heterogeneous raw datasets.

[0008] The multi-dimensional parameter adaptive fusion unit receives the raw dataset output by the multi-source parameter acquisition unit, incorporates an adaptive confidence model based on improved Kalman filtering, and completes data time dimension calibration by configuring a constraint confidence time window for each type of parameter and combining time decay confidence weights and operating condition reliability factors. Simultaneously, it achieves noise filtering of multi-source heterogeneous data through physical constraint calibration, introduces electromagnetic environment dimension calibration, calculates the electromagnetic attenuation coefficient based on UWB signal strength, phase difference, and multipath delay, dynamically adjusts the contribution of UWB data in the fusion weights, and finally completes the standardized fusion of multi-source data, outputting a high-confidence fused sensing dataset.

[0009] The process node dynamic parsing unit pre-stores the standard process flow chart, matches and fuses the perception data to determine the current work node, combines the resource status to optimize the node threshold, and outputs the process parsing result with constraints.

[0010] The intelligent decision-making and instruction generation unit combines a gradient boosting tree model and a work rule engine, inputs parsing results and key parameters to generate work instructions, and after verification, converts them into industrial protocol signals and sends them to the equipment controller.

[0011] The closed-loop control and verification unit collects instruction execution feedback data, compares it with preset indicators, generates a correction signal if the deviation exceeds the threshold, and optimizes the collection frequency, confidence weight, node threshold and model parameters in reverse.

[0012] As a further improvement to this technical solution, the multi-source parameter acquisition unit includes an industrial bus acquisition module, a distributed sensing module, a machine vision and NFC acquisition module, a UWB positioning acquisition module, and a cargo handling and shift data acquisition module, wherein:

[0013] The industrial bus acquisition module is based on the sensor interface of the dock operation equipment and adopts industrial bus communication technology to collect the operating parameters of the equipment.

[0014] The distributed sensing module integrates environmental sensors and weighbridge equipment to collect data on ambient temperature and humidity, wind speed, visibility, and cargo weighing.

[0015] The machine vision and NFC acquisition module acquires work ticket number and work line status information based on machine vision recognition technology, and binds cargo area, cargo location, stack number and cargo type and packaging type information with the help of NFC technology.

[0016] The UWB positioning acquisition module uses UWB positioning technology to track the real-time trajectory data of workers and vehicles.

[0017] The cargo handling and shift data acquisition module synchronously acquires work shift information, collects core cargo handling data, and summarizes the collected data to form a multi-source heterogeneous raw dataset.

[0018] As a further improvement to this technical solution, the multi-dimensional parameter adaptive fusion unit includes a time dimension calibration module, a physical constraint noise reduction module, an electromagnetic environment calibration module, and a standardized fusion output module, wherein:

[0019] The time dimension calibration module receives the multi-source heterogeneous raw dataset output by the multi-source parameter acquisition unit and calls the adaptive confidence model to perform time dimension calibration.

[0020] The physical constraint noise reduction module interfaces with the output data of the time dimension calibration module to perform noise filtering of multi-source heterogeneous data.

[0021] The electromagnetic environment calibration module receives the output data from the UWB positioning and acquisition module and performs electromagnetic dimension calibration of the UWB data.

[0022] The standardized fusion output module connects with the processing results of the time dimension calibration module, the physical constraint noise reduction module, and the electromagnetic environment calibration module to complete the standardized fusion of multi-source data and output a high-confidence fusion sensing dataset.

[0023] As a further improvement to this technical solution, the calibration process of the time dimension calibration module includes the following steps:

[0024] S21.1 Based on the parameter types and operational accuracy requirements of multi-source heterogeneous raw datasets, the reliable time window of the constraint is calibrated through on-site testing at the dock, the window of the equipment operating parameters is adapted to the equipment response speed, and the window of the environment and weighing data is adapted to the data stability requirements.

[0025] S21.2, Calculate the time decay confidence weights using the adaptive confidence model. , Data collection time difference Working condition adaptability coefficient Relatedness, fit coefficient Calibration was performed through actual measurements in real-world dock operation scenarios.

[0026] S21.3, Apply time decay confidence weights Multiplying the corresponding parameter by the operating condition reliability factor yields the time dimension confidence weight of the parameter. The operating condition reliability factor is calibrated based on the statistical results of data credibility under different operating conditions at the dock.

[0027] As a further improvement to this technical solution, the calibration process of the electromagnetic environment calibration module includes the following steps:

[0028] S23.1 Receive the UWB signal strength output by the UWB positioning acquisition module Phase difference and multipath delay Actual measurement data;

[0029] S23.2 Calculate the electromagnetic attenuation coefficient based on the data collected in S23.1 , and , , It is related to the scene adaptation weight coefficient, and the weight coefficient is measured and calibrated according to the distribution of interference sources at the dock and the operation area classification.

[0030] S23.3, Establish A correlation model with UWB data distortion is used to dynamically adjust the confidence weights of the time dimension of UWB data based on the correlation model. The contribution of UWB data fusion varies with... Adaptive adjustment to changes.

[0031] As a further improvement to this technical solution, the process of standardizing and fusing multi-source data and outputting a high-confidence fused sensing dataset by the standardized fusion output module includes the following steps:

[0032] S24.1, synchronously receive the parameters with time-dimensional confidence weights output by the time-dimensional calibration module, the denoised data output by the physical constraint denoising module, and the data output by the electromagnetic environment calibration module. Adjusted UWB data;

[0033] S24.2. Based on the time dimension confidence weights of each parameter, perform weighted fusion calculations on data of the same dimension. The fusion result is then compared with the parameter confidence weights. And directly related to the measured data;

[0034] S24.3. The min-max normalization method is used to process the fused data of all dimensions to form a standardized high-confidence fused perception dataset.

[0035] As a further improvement to this technical solution, the process node dynamic parsing unit includes a graph storage module, a node matching module, a threshold optimization module, and a parsing result output module, wherein:

[0036] The graph storage module is used to pre-store standard process graphs of operations classified by operation type. The operation types cover scenarios such as loading and unloading ships, loading and unloading trains, loading and unloading automobiles at ports, handling and repackaging. The graphs contain key nodes and node feature parameters corresponding to each scenario.

[0037] The node matching module receives the high-confidence fusion perception dataset output by the multi-dimensional parameter adaptive fusion unit, extracts key parameters, and matches them with the node feature parameters of the standard workflow graph to determine the current job node.

[0038] The threshold optimization module obtains the status data of dock operation resources and optimizes the judgment threshold of the current operation node in combination with the status data of dock operation resources.

[0039] The parsing result output module integrates the current job node information with the optimized threshold, and outputs the process parsing result including job priority and resource constraints.

[0040] As a further improvement to this technical solution, the intelligent decision-making and instruction generation unit includes a model and rule configuration module, a job instruction generation module, an instruction verification module, and a protocol conversion and transmission module, wherein:

[0041] The model and rule configuration module is used to store the trained gradient boosting tree model and configure the rigid rules related to dock operations.

[0042] The job instruction generation module receives the process parsing results output by the process node dynamic parsing unit, and generates preliminary job instructions by combining key job parameters and through the collaborative operation of the gradient boosting tree model and the job rule engine.

[0043] The instruction verification module performs security threshold verification and permission verification on the initial operation instructions.

[0044] The protocol conversion and transmission module converts the verified operation instructions into industrial protocol signals and sends them to the dock operation equipment controller.

[0045] As a further improvement to this technical solution, the configuration content of the model and rule configuration module specifically includes:

[0046] The training data for the gradient boosting tree model comes from historical terminal operation data such as equipment operation thresholds, cargo space occupancy rates, and operation duration statistics.

[0047] The configuration rules of the job rule engine cover safe operation specifications, job priority rules, and equipment operation permission rules;

[0048] The collaborative logic between the gradient boosting tree model and the job rule engine is as follows: the gradient boosting tree model outputs dynamic job decision suggestions, the rule engine performs rigid constraint verification on the suggestions, and the gradient boosting tree model and the job rule engine work together to output decision results that meet the needs of the scenario, providing a basis for the generation of job instructions.

[0049] As a further improvement to this technical solution, the closed-loop control and verification unit includes a feedback data acquisition module, an index comparison module, a correction signal generation module, and a parameter reverse optimization module, wherein:

[0050] The feedback data acquisition module is used to collect instruction execution feedback data of the dock operation equipment, including operation completion rate, actual parameter execution value, and equipment operation status feedback data.

[0051] The indicator comparison module compares the feedback data with preset operation indicators and calculates the data deviation value.

[0052] The correction signal generation module is used to generate a targeted correction signal when the deviation value exceeds a set threshold.

[0053] The parameter reverse optimization module transmits the correction signal to the multi-source parameter acquisition unit, the multi-dimensional parameter adaptive fusion unit, the process node dynamic analysis unit, and the intelligent decision and instruction generation unit, respectively, to reverse optimize the acquisition frequency of the multi-source parameter acquisition unit, the confidence weight of the multi-dimensional parameter adaptive fusion unit, the node threshold of the process node dynamic analysis unit, and the model parameters of the intelligent decision and instruction generation unit.

[0054] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0055] 1. This invention integrates industrial bus, distributed sensing, machine vision, UWB positioning, and NFC technology through a multi-source parameter acquisition unit to comprehensively collect multi-dimensional data such as equipment operation, environmental weighing, cargo information, personnel and vehicle trajectories, and cargo handling shifts. Simultaneously, the multi-dimensional parameter adaptive fusion unit incorporates an adaptive confidence model based on improved Kalman filtering, configuring constrained reliable time windows for various parameters. It combines time decay confidence weights and operational reliability factors to complete time dimension calibration, further supplemented by physical constraint noise reduction and electromagnetic environment calibration, achieving accurate fusion of multi-source heterogeneous data. Finally, the process node dynamic parsing unit, through pre-stored standard process diagrams for various operations such as ship loading / unloading, train loading / unloading, and vehicle port access, completes accurate adaptation across multiple scenarios, constructing an integrated management and control architecture covering the entire process of bulk cargo terminal operations, thus overcoming the shortcomings of existing technologies that do not cover all business scenarios.

[0056] 2. This invention addresses the technical problem of insufficient data governance and closed-loop management by using a multi-dimensional parameter adaptive fusion unit with a min-max standardization method to complete the unified processing of multi-source data, establishing a standardized data fusion standard and ensuring data consistency. Simultaneously, the closed-loop management and verification unit collects instruction execution feedback data, compares it with preset indicators, generates correction signals, and reverse-optimizes the collection frequency, confidence weight, node thresholds, and model parameters, forming a full-process data closed-loop mechanism of "collection-fusion-analysis-decision-feedback-optimization," meeting the needs of refined management of bulk cargo terminals for controllable data quality and traceable responsibility.

[0057] 3. This invention addresses the technical problems of insufficient full-chain intelligence and multi-terminal collaboration. By combining an intelligent decision-making and instruction generation unit with a gradient boosting tree model (trained based on historical operation data) and an operation rule engine, it achieves intelligent generation and safety verification of operation instructions, promoting intelligent decision-making. The multi-source parameter acquisition unit supports collaborative data acquisition using multiple technologies such as industrial bus, UWB positioning, NFC, and machine vision, adapting to the data interaction needs of different terminals, reducing manual intervention, and improving the intelligence and efficiency of bulk cargo terminal operations. This solves the problems of insufficient intelligence and lack of multi-terminal collaboration in existing technologies. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the system framework of the present invention;

[0059] The meanings of the labels in the diagram are as follows:

[0060] 1. Multi-source parameter acquisition unit; 11. Industrial bus acquisition module; 12. Distributed sensing module; 13. Machine vision and NFC acquisition module; 14. UWB positioning acquisition module; 15. Cargo handling and shift data acquisition module;

[0061] 2. Multi-dimensional parameter adaptive fusion unit; 21. Time dimension calibration module; 22. Physical constraint noise reduction module; 23. Electromagnetic environment calibration module; 24. Standardized fusion output module;

[0062] 3. Dynamic parsing unit for process nodes; 31. Graph storage module; 32. Node matching module; 33. Threshold optimization module; 34. Parsing result output module;

[0063] 4. Intelligent decision-making and instruction generation unit; 41. Model and rule configuration module; 42. Job instruction generation module; 43. Instruction verification module; 44. Protocol conversion and transmission module;

[0064] 5. Closed-loop control and verification unit; 51. Feedback data acquisition module; 52. Indicator comparison module; 53. Correction signal generation module; 54. Parameter reverse optimization module. Detailed Implementation

[0065] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0066] like Figure 1 As shown, this embodiment provides a multi-dimensional parameter perception-based end-to-end control system for bulk cargo terminal operations, including:

[0067] Multi-source parameter acquisition unit 1 adopts a multi-source architecture that combines industrial bus, distributed sensing, machine vision, UWB positioning and NFC technology to collect equipment operating parameters, environmental and weighing data, cargo information, operator and vehicle trajectories, cargo handling and shift data, and output multi-source heterogeneous raw datasets.

[0068] In this embodiment, the multi-source parameter acquisition unit 1 includes an industrial bus acquisition module 11, a distributed sensing module 12, a machine vision and NFC acquisition module 13, a UWB positioning acquisition module 14, and a sorting and shift data acquisition module 15, wherein:

[0069] The industrial bus acquisition module 11 is based on the sensor interface of the dock operation equipment and adopts industrial bus communication technology to acquire the operating parameters of the equipment.

[0070] Specifically, the industrial bus acquisition module 11 adopts the Profinet and ModbusTCP dual-protocol architecture commonly used in dock equipment. It is compatible with the analog and digital sensor interfaces of core and auxiliary equipment such as cranes, belt conveyors, and stacker-reclaimers. It focuses on acquiring core operating parameters of the equipment, such as lifting weight, speed, current, temperature, displacement, and pressure. The acquisition frequency is dynamically adjusted according to the operating status of the equipment (low-frequency acquisition under steady-state conditions, and high-frequency acquisition under dynamic conditions such as start-up, shutdown, and load changes). Real-time data transmission is achieved through standard industrial communication protocols to ensure compatibility with various dock operation equipment and data transmission stability.

[0071] The distributed sensing module 12 integrates environmental sensors and weighbridge equipment to collect data on ambient temperature and humidity, wind speed, visibility, and cargo weighing.

[0072] Specifically, environmental sensors are deployed at the four corners of the work area, the highest point of the yard, and the top of the equipment. Weighbridges are deployed at the dock entrance, yard entrance and exit, and the end of the work line based on the dock operation equipment. Environmental data is collected at fixed intervals, and weighing data is collected by weight triggering and automatically associated with vehicle or cargo identification information. All collected data is transmitted to the control host after being in a unified format to ensure the correlation and effectiveness of environmental data and weighing data, providing basic data support for judging the working conditions.

[0073] The machine vision and NFC acquisition module 13 acquires work ticket number and work line status information based on machine vision recognition technology, and binds cargo area, cargo location, stack number and cargo type and packaging type information with the help of NFC technology.

[0074] Specifically, industrial cameras and supplementary lights are installed at key locations at the entrance of the work line and in the cargo area. OCR technology is adapted to the outdoor environment of the dock to accurately identify the work ticket number and the operation / stagnation / fault status of the work line. At the same time, NFC fixed reading and writing terminals are deployed in the cargo area and next to the cargo location, as well as handheld terminals carried by the operators. When goods enter the warehouse, the terminal binds the cargo area, cargo location, stack number and cargo type and packaging type information. When goods leave the warehouse or are transferred, the terminal quickly identifies the associated information. The module supports offline data caching function and automatically synchronizes after the network is restored to avoid data loss.

[0075] The UWB positioning acquisition module 14 uses UWB positioning technology to track the real-time trajectory data of workers and vehicles.

[0076] Specifically, UWB base stations are deployed on pillars and building walls in the bulk cargo terminal operation area according to the triangle networking principle to ensure full coverage of the operation area. Workers integrate positioning tags into their safety helmets, and work vehicles fix positioning tags to the top of the cab. The tags are bound to the employee's number and the vehicle's license plate number respectively. During the data collection process, the positioning refresh rate is set differently according to the movement speed of personnel and vehicles. After receiving the tag signal, the base station transmits it to the control host, calculates the three-dimensional coordinate data, and finally outputs trajectory data containing identity identifiers, timestamps, and real-time coordinates.

[0077] The cargo handling and shift data acquisition module 15 synchronously acquires work shift information, collects core cargo handling data, and summarizes the collected data to form a multi-source heterogeneous raw dataset.

[0078] Specifically, the cargo handling and shift data acquisition module 15 connects to the existing attendance management system of the bulk cargo terminal through a standard interface to obtain shift information such as shift number, start and end time, work group, and responsible area. At the same time, it uses handheld terminals or scanning devices to collect core cargo handling data such as cargo name, quantity, receiving and dispatching units, and means of transport number (ship name, license plate number, train car number). When aggregating valid data collected by the industrial bus acquisition module 11, distributed sensing module 12, machine vision and NFC acquisition module 13, and UWB positioning acquisition module 14, it automatically removes invalid data such as timeouts and incorrect formats, and encapsulates it in a unified format of "module identifier-parameter type-timestamp-data value" to finally form a standardized multi-source heterogeneous raw dataset.

[0079] The multi-dimensional parameter adaptive fusion unit 2 receives the raw dataset output by the multi-source parameter acquisition unit 1. It incorporates an adaptive confidence model based on improved Kalman filtering. By configuring a constraint confidence time window for each type of parameter, it combines time decay confidence weights and operating condition reliability factors to complete the data time dimension calibration. Simultaneously, it achieves noise filtering of multi-source heterogeneous data through physical constraint calibration and introduces electromagnetic environment dimension calibration. Based on UWB signal strength, phase difference, and multipath delay, it calculates the electromagnetic attenuation coefficient and dynamically adjusts the contribution of UWB data in the fusion weight. Finally, it completes the standardized fusion of multi-source data and outputs a high-confidence fused sensing dataset. The multi-dimensional parameter adaptive fusion unit 2 implements its core functions based on an industrial-grade embedded computing platform (equipped with a Linux operating system, integrating a multi-core processor and high-speed cache, and supporting parallel data processing). It establishes communication with the multi-source parameter acquisition unit 1 through a standardized data interface and receives multi-source heterogeneous raw datasets in real time. The multi-dimensional parameter adaptive fusion unit 2 incorporates an adaptive confidence model based on an improved Kalman filter. The improvement lies in integrating time-decay confidence weights and operational reliability factors into the state update equation of the Kalman filter, enhancing data calibration accuracy under dynamic operating conditions. The model runs on the computing platform's algorithm engine (written in C++, supporting CUDA acceleration), supporting dynamic parameter configuration and real-time invocation. Each functional module executes in a serial flow: "time dimension calibration → physical constraint noise reduction → electromagnetic environment calibration → standardized fusion." Data is transferred between modules via shared memory (SHM) (transfer rate ≥ 1GB / s) to avoid data copy latency. The computing platform has a built-in data cache (cache capacity ≥ 10GB) and log recording function (log format: "timestamp-module ID-data identifier-processing status"), retaining processing data for each module (retention period 7 days) for fault tracing and parameter optimization. A watchdog timer (timeout threshold 3s) is also configured to prevent module freezes and process interruptions. The final high-confidence fusion sensing dataset is transmitted in real time to the process node dynamic parsing unit 3 via a gigabit Ethernet interface, with a transmission rate of ≥500Mbps and support for resuming interrupted transmission.

[0080] The multi-dimensional parameter adaptive fusion unit 2 includes a time dimension calibration module 21, a physical constraint noise reduction module 22, an electromagnetic environment calibration module 23, and a standardized fusion output module 24, wherein:

[0081] In this embodiment, the time dimension calibration module 21 receives the multi-source heterogeneous raw dataset output by the multi-source parameter acquisition unit 1 and calls the adaptive confidence model to perform time dimension calibration; the calibration process of the time dimension calibration module 21 includes the following steps:

[0082] S21.1 Based on the parameter types and operational accuracy requirements of multi-source heterogeneous raw datasets, the reliable time window of the constraint is calibrated through on-site testing at the dock, the window of the equipment operating parameters is adapted to the equipment response speed, and the window of the environment and weighing data is adapted to the data stability requirements.

[0083] Specifically, based on the parameter types and operational accuracy requirements of the multi-source heterogeneous raw datasets, calibration was performed using field testing. For equipment operating parameters, calibration was categorized according to differences in equipment response speed. For equipment with faster response speeds (such as cranes and stacker-reclaimers), the calibration window was adapted to its millisecond-level action response characteristics; for equipment operating in a steady state (such as belt conveyors and conveyor pumps), the calibration window was adapted to its stable data output characteristics. For environmental and weighing data, the calibration window was adapted to the stability requirements of natural data fluctuations. The testing process covered different operating periods and load states at the bulk cargo terminal, recording the reliability changes of parameters within different time intervals. The longest time interval during which each type of parameter maintained effective reliability was determined as the constraint reliability time window.

[0084] S21.2, Calculate the time decay confidence weights using the adaptive confidence model. , Data collection time difference Working condition adaptability coefficient Relatedness, fit coefficient Calibration was performed based on actual operational scenarios at bulk cargo terminals.

[0085] To improve the accuracy of data confidence weight calculation, the adaptive confidence model in this embodiment is based on an improved Kalman filter, as detailed below:

[0086] The original Kalman filter core equation:

[0087] Equations of state: ;in for Time-parameter state vector Here is the state transition matrix. This is process noise;

[0088] Observation equation: ;in for Observation vector at time, For the observation matrix, To observe noise.

[0089] Improved state update equation: fused constraint reliable time window Working condition adaptability coefficient The improved state prior estimation equation is as follows:

[0090] ;

[0091] in for Time-prior state estimation Due to the time difference in data collection, This is to correct for process noise.

[0092] Specifically, the adaptive confidence model is invoked, based on the data collection time difference. The relationship with the constrained credible time window is calculated using a piecewise nonlinear formula. The core formula is as follows:

[0093] when ( When constraining the credible time window (i.e., the constrained credible time window in S21.1):

[0094] ;

[0095] when hour:

[0096] ;

[0097] in, The time decay confidence weight (dimensionless, value range [0,1]); This refers to the data acquisition time difference, which is the difference between the data processing time and the acquisition time. To constrain the reliable time window (calibrated by S21.1); The working condition adaptation coefficient (dimensionless) is determined by actual measurement and calibration in the actual operation scenarios of bulk cargo terminals. It is classified according to core working conditions such as loading and unloading ships, loading and unloading trains, truck collection and distribution, handling, and repackaging, to adapt to the data attenuation characteristics under different working conditions.

[0098] S21.3, Apply time decay confidence weights Multiplying the corresponding parameter by the operating condition reliability factor yields the time dimension confidence weight of the parameter. The operating condition reliability factor is calibrated based on the statistical results of data credibility under different operating conditions at the dock.

[0099] Specifically, a working condition reliability factor is introduced. This factor is calibrated based on the historical data reliability statistics under different operating conditions at the bulk cargo terminal. The statistical logic is the ratio of the amount of valid data to the total amount of data under each operating condition. Valid data refers to data that meets physical constraints and has no abnormal deviations. The final time dimension confidence weight is determined accordingly. The calculation formula is as follows:

[0100] ;

[0101] The definitions of each symbol are as follows: The time dimension confidence weight for the parameter (dimensionless, value range [0,1]); The time decay confidence weight (calculated from S21.2); The operating condition reliability factor (dimensionless).

[0102] when If the data falls below the set low confidence threshold, it is marked as temporarily stored for subsequent cross-validation.

[0103] In this embodiment, the physical constraint noise reduction module 22 interfaces with the output data of the time dimension calibration module 21 to perform noise filtering of multi-source heterogeneous data;

[0104] Specifically, the physical constraint noise reduction module 22 connects to the output data of the time dimension calibration module 21, and constructs a constraint rule base based on the physical characteristics of the parameters, equipment technical specifications, and industry standards to perform noise filtering of multi-source heterogeneous data. The specific implementation is as follows:

[0105] The constraint rule base is constructed according to parameter type. The rule content covers the reasonable value range of parameters, and the threshold sources include technical manuals for bulk cargo terminal operation equipment, relevant industry safety regulations, and natural characteristics of environmental parameters.

[0106] The noise filtering process is as follows:

[0107] First, the input data is compared with the corresponding constraint rules to identify outliers that exceed the reasonable range. For a single outlier, a moving average filter is used for correction, as shown in the following formula:

[0108] ;

[0109] in, Correction values ​​for outlier data; The number of consecutive valid data sets before and after the abnormal data is the size of the sliding window. For the first in the sliding window The group contains valid data that conforms to physical constraints.

[0110] If multiple sets of data exceed the constraint range consecutively, they are determined to be invalid data, triggering an alarm signal and feeding back to the multi-source parameter acquisition unit 1. At the same time, the historical valid data of the parameter is temporarily used and the status is marked.

[0111] In this embodiment, the electromagnetic environment calibration module 23 receives the output data from the UWB positioning and acquisition module 14 and performs electromagnetic dimension calibration of the UWB data; the calibration process of the electromagnetic environment calibration module 23 includes the following steps:

[0112] S23.1 Receive the UWB signal strength output by the UWB positioning acquisition module 14 Phase difference and multipath delay Measured data; After receiving the above data, the format is first checked to confirm that it contains key fields such as parameter name, collection timestamp, data value, and validity identifier, and data with incorrect format or missing fields is removed; then the validity is checked according to the effective communication range of the UWB device and the physical value range of the parameters, and the valid data that meets the requirements is selected and sorted by collection timestamp to ensure time sequence consistency.

[0113] S23.2 Calculate the electromagnetic attenuation coefficient based on the data collected in S23.1 , and , , It is related to the scene adaptation weight coefficient, and the weight coefficient is measured and calibrated according to the distribution of interference sources in the bulk cargo terminal and the operation area classification.

[0114] Specifically, based on the received measured data and scenario adaptation weight coefficients, a weighted summation formula is used to calculate... The core formula is as follows:

[0115] ;

[0116] in, UWB signal strength (unit: dBm) Its standardized value is calculated as follows: ;in , The measured extreme values ​​of UWB signal strength in the general cargo terminal operation scenario;

[0117] The phase difference of the UWB signal (unit: rad). Its standardized value is calculated as follows: (In the dock scenario, the phase difference ranges from [0, ...) ]);

[0118] Multipath delay of UWB signal (unit: ns). Its standardized value is calculated as follows: ;in This represents the measured maximum value of UWB multipath delay in a bulk cargo terminal operation scenario;

[0119] To adapt the weight coefficients to the scene (dimensionless, satisfying...) According to the distribution of interference sources in bulk cargo terminals (such as the working areas of hold side and yard), the electromagnetic interference intensity and UWB data distortion of each area are collected and the weight coefficients of the corresponding areas are determined by linear fitting.

[0120] S23.3, Establish A correlation model with UWB data distortion is used to dynamically adjust the confidence weights of the time dimension of UWB data based on the correlation model. The contribution of UWB data fusion varies with... Adaptive adjustment to changes.

[0121] Specifically, first establish UWB data distortion The association model, Defined as the deviation rate between UWB positioning data and reference positioning data, the correlation model is obtained by collecting sufficient data under different electromagnetic interference environments and using quadratic polynomial fitting.

[0122] Subsequently, the confidence weights of the time dimension of the UWB data are dynamically adjusted based on this model, using the following formula:

[0123] ;

[0124] in, Adjust the confidence weights for the time dimension of the UWB data; The time dimension confidence weight of the UWB data before adjustment (calculated by the time dimension calibration module 21); The distortion attenuation coefficient is based on... Size segmentation settings, The larger the value, the smaller the attenuation coefficient, ensuring that the contribution of UWB data fusion is adaptively reduced under high interference environments.

[0125] In this embodiment, the standardized fusion output module 24 receives the processing results from the time dimension calibration module 21, the physical constraint noise reduction module 22, and the electromagnetic environment calibration module 23 to complete the standardized fusion of multi-source data and output a high-confidence fused sensing dataset. The process of standardized fusion of multi-source data and outputting a high-confidence fused sensing dataset by the standardized fusion output module 24 includes the following steps:

[0126] S24.1, synchronously receive the parameters with time-dimension confidence weights output by the time-dimension calibration module 21, the denoised data output by the physical constraint denoising module 22, and the data output by the electromagnetic environment calibration module 23. Adjusted UWB data;

[0127] Specifically, three types of data are received synchronously, and data alignment is achieved based on the collection timestamp, with alignment accuracy controlled within a reasonable range. A data receiving buffer is set up. For delayed data that has not been received within a time limit, key parameters are temporarily replaced by valid data from the previous time window and their status is marked. Non-key parameters are directly marked as missing data. After reception, integrity verification is performed to ensure that each data entry contains key information such as parameter identifier, timestamp, data value, and confidence weight. Data lacking key information is directly discarded.

[0128] S24.2. Based on the time dimension confidence weights of each parameter, perform weighted fusion calculations on data of the same dimension. The fusion result is then compared with the parameter confidence weights. And directly related to the measured data;

[0129] Specifically, based on parameter dimensions (equipment operating parameters, environmental data, UWB positioning data, etc.), weighted fusion calculations are performed on valid data within the same dimension. The core formula is as follows:

[0130] ;

[0131] in, This is the result of data fusion within the same dimension; This represents the number of valid data points within the same dimension (excluding low-confidence data with confidence weights below a set threshold). For the first Confidence weights for each data point (UWB data uses adjusted weights). The remaining data are weighted using the time dimension. ); For the first The measured or corrected value of each data point.

[0132] After fusion, the results are verified. If the mean deviation from the valid input data exceeds a reasonable range, the fusion calculation is performed again.

[0133] S24.3. The min-max normalization method is used to process the fused data of all dimensions to form a standardized high-confidence fused perception dataset.

[0134] Specifically, the min-max standardization method is used to uniformly process the fused data across all dimensions, eliminating differences in units. The core formula is as follows:

[0135] ;

[0136] in, The standardized parameter values ​​(dimensionless, ranging from [0,1]); This is the result of data fusion within the same dimension; This is the historical minimum value of this dimension parameter (determined based on historical operational data statistics of the bulk cargo terminal). This is the historical maximum value of this dimension parameter (determined based on historical operational data statistics of the bulk cargo terminal).

[0137] If a certain dimension parameter has no fluctuation ( If the data is not standardized, its standardized value is set to a fixed intermediate value. After standardization, the data is packaged in the format of "parameter dimension - collection timestamp - standard value - confidence weight - data status" to form a high-confidence fusion perception dataset. This dataset is then output to the process node dynamic parsing unit 3 in real time through a standard communication interface, while also being backed up locally.

[0138] It should be added that, regarding the constraint credible time window Working condition adaptability coefficient Scene adaptation weight coefficient The calibration method is as follows:

[0139] Calibration equipment selection: UWB positioning base stations, industrial-grade temperature and humidity sensors, crane load sensors and data acquisition instruments (model: NICDAQ-9178) commonly used in bulk cargo terminal operations are adopted.

[0140] Constrained Confidential Time Window Calibration:

[0141] Test procedure: Select three typical operating areas (hull side, yard, and gate) of the bulk cargo terminal, and continuously collect the transmission delay of UWB positioning data 100 times in each area;

[0142] Data recording period: Delay data is recorded once every 200ms;

[0143] Valid data judgment criteria: Remove abnormal data with a delay > 100ms (accounting for ≤ 5%).

[0144] Statistical method: The 95th percentile lag value of the valid data is taken as... Example result: Cabinside area =40ms, storage area =50ms.

[0145] Operating condition adaptability coefficient Calibration:

[0146] Operating conditions covered: Includes three typical operating conditions: "sunny day - light load", "rainy day - heavy load" and "night - medium load", with each condition tested 5 times.

[0147] Test duration for each working condition: Continuous data collection for 2 hours;

[0148] Data statistical method: Calculate the mean deviation rate between actual parameter values ​​and fused values ​​for each type of working condition using the following formula: Mean deviation rate, example result: "Sunny-light load" condition =0.92, "Rainy Day - Heavy Load" Working Condition 0.85.

[0149] Scene adaptation weight coefficient Calibration:

[0150] Test sample size: 500 sets of data were collected for each scenario (UWB positioning, visual recognition, device status);

[0151] Statistical method: The analytic hierarchy process (AHP) was used to calculate the reliability score for each scenario, and weights were assigned according to the score proportions. Example results: =0.4 (UWB positioning) =0.35 (visual recognition) =0.25 (device status).

[0152] Calibration result verification method: Substitute the calibration parameters into the system and run it continuously for 48 hours. If the deviation rate of the operation data is ≤5%, the calibration result is valid.

[0153] The process node dynamic analysis unit 3 has a pre-stored standard process flow chart, matches and integrates the perception data to determine the current operation node, combines the resource status to optimize the node threshold, and outputs the process analysis result with constraints.

[0154] In this embodiment, the process node dynamic parsing unit 3 includes a graph storage module 31, a node matching module 32, a threshold optimization module 33, and a parsing result output module 34, wherein:

[0155] The graph storage module 31 is used to pre-store the standard process graphs of operations classified by operation type. The operation types cover scenarios such as loading and unloading ships, loading and unloading trains, automobile collection and distribution at ports, handling and repackaging. The graphs contain key nodes and node feature parameters corresponding to each scenario.

[0156] Specifically, the graph storage module 31 is used to pre-store standard process graphs categorized by job type. These job types cover five core scenarios: ship loading / unloading, train loading / unloading, port vehicle transport, handling, and repackaging. Each job type corresponds to an independent directed graph, and each graph contains complete key nodes for the corresponding scenario, along with the unique feature parameters of each node.

[0157] Key nodes are divided into start nodes, intermediate nodes and end nodes according to the order of the operation process. The node characteristic parameters are configured according to the different scenarios. For example, the nodes in the loading and unloading scenario include parameters such as ship identification, berthing status and manifest verification completeness. The nodes in the automobile collection and distribution port scenario include parameters such as vehicle identification, cargo bill information completeness and weighing data validity.

[0158] Furthermore, the graph storage module 31 adopts a hybrid storage architecture of "relational database + graph database". The basic information of nodes (including node ID, name, type, benchmark judgment threshold, feature parameter list, etc.) is stored in the MySQL database. The relationship between nodes is stored in the Neo4j graph database in the form of triples of "preceding node ID-triggering condition-subsequent node ID". This not only ensures the structured management of node information, but also optimizes the query efficiency of the flow logic. It also supports two maintenance modes: manual update and automatic update. Manual update requires a "submission-review-effectiveness" process, while automatic update is triggered based on the long-term statistical regularity of node triggering conditions. All update operations are recorded in complete logs to ensure the traceability and accuracy of graph data.

[0159] The node matching module 32 receives the high-confidence fusion perception dataset output by the multi-dimensional parameter adaptive fusion unit 2, extracts key parameters, and matches them with the node feature parameters of the standard workflow graph to determine the current work node.

[0160] Specifically, after the node matching module 32 receives the high-confidence fused sensing dataset output by the multi-dimensional parameter adaptive fusion unit 2:

[0161] First, the current job type is quickly identified by the job order number prefix, and then the core parameters for node determination in this scenario are extracted according to the preset parameter filtering rules.

[0162] The extracted parameters were then standardized. Boolean parameters were uniformly converted to "1 (satisfied) / 0 (not satisfied)", numerical parameters retained their original 0-1 normalized values, and text parameters were converted to 32-bit MD5 hash values. Invalid parameters with too low confidence weights or those exceeding the reasonable range were also removed.

[0163] Next, a weighted matching algorithm is used to calculate the matching degree between the fused data and each node. The matching degree is obtained by the sum of the products of the fit of each feature parameter and the corresponding weight. The feature parameter weights are marked according to their importance to the node judgment and the sum is 1. The fit is calculated differently according to the parameter type. Boolean and text parameters are judged by "1 if consistent and 0 if inconsistent". Numerical parameters are calculated according to the deviation ratio within the reasonable fluctuation range of the standard value, and 0 if they exceed the range.

[0164] Finally, the judgment logic is executed based on the matching degree result. If there is a single node with a matching degree greater than or equal to the node baseline threshold, it is directly judged as the current operation node. If the matching degree of all nodes does not meet the baseline threshold, a transition probability matrix is ​​constructed based on the historical operation node flow sequence to filter candidate nodes with high prediction probability. If the candidate node is unique, it is marked as "pending review". Otherwise, a manual confirmation process is triggered to ensure the accuracy of the current operation node judgment.

[0165] The threshold optimization module 33 acquires the status data of dock operation resources and optimizes the judgment threshold of the current operation node in combination with the status data of dock operation resources.

[0166] Specifically, the process for optimizing the judgment threshold of the current job node is as follows:

[0167] The threshold optimization module 33 first connects to the bulk cargo terminal resource management system through the standard communication protocol to obtain real-time resource status data of the bulk cargo terminal operation. The collection scope covers four core resources: equipment resources (total quantity, occupied quantity, number of faults), human resources (number of on-duty personnel, number of qualified personnel), cargo yard resources (total number of cargo spaces, number of available cargo spaces, congestion status), and transportation resources (waiting time for scheduling, preset maximum allowable waiting time). The collection frequency is consistent with the output frequency of the high confidence fusion perception dataset. The collected data is cached in local memory and managed according to the first-in-first-out principle.

[0168] Subsequently, based on the collected resource status data, the saturation of individual resources is calculated. The saturation of equipment, personnel, cargo yard, and transportation is calculated by "number of occupied equipment / total number of equipment", "number of on-duty personnel with matching qualifications / number of personnel required for operation", "1 - number of available cargo spaces in the target cargo area / total number of cargo spaces", and "waiting time for scheduling / preset maximum allowable waiting time". The comprehensive resource saturation is then obtained by weighted summation. The weights of various resources are statistically calibrated according to the actual working conditions of the terminal.

[0169] Finally, the optimization coefficient is determined based on the overall resource saturation. When resources are sufficient, the optimization coefficient is set to 0.9 (threshold lowered), when resources are balanced, it is set to 1.0 (threshold unchanged), and when resources are scarce, it is set to 1.1 (threshold higher). The actual judgment threshold of the current node is calculated by "adjusted threshold = node baseline judgment threshold × optimization coefficient". The adjusted threshold is only valid for the current node. At the same time, if the matching degree fluctuates too much after multiple consecutive optimizations, the optimization coefficient is automatically locked and an alarm is triggered to ensure the stability and rationality of the threshold adjustment.

[0170] The parsing result output module 34 integrates the current job node information and the optimized threshold to output the process parsing result including job priority and resource constraints.

[0171] Specifically, the parsing result output module 34 first integrates the data processed by each module to form structured data containing four dimensions: basic operation information, current node information, threshold optimization information, and constraint information. The basic operation information includes the operation order number, operation type, initiation time, and associated transport vehicle identifier; the current node information includes the node ID, name, matching degree, and node status; and the threshold optimization information includes the baseline judgment threshold, comprehensive resource saturation, optimization coefficient, and adjusted threshold. Then, the constraint information is clarified. The operation priority is calculated using a weighted scoring method, with scoring indicators including cargo urgency, transport vehicle waiting time weight, etc. Resource matching is categorized into high, medium, and low levels based on the scoring results. Resource constraints clearly define the type, quantity, operating status, number of personnel, and qualification requirements for each node. Environmental and time constraints are set based on actual operational needs, including reasonable ranges of environmental parameters, maximum operating time, and prohibited operating periods. Finally, the integrated process analysis results are output in a standardized JSON format and transmitted via Gigabit Ethernet using the TCP protocol to the intelligent decision-making and instruction generation unit 4. A data verification mechanism is enabled during transmission. If transmission fails, a retransmission operation is automatically performed. If retransmission fails, an alarm is triggered, and the analysis results are backed up locally to ensure the integrity and reliability of data transmission.

[0172] The intelligent decision-making and instruction generation unit 4 combines the gradient boosting tree model and the operation rule engine, inputs the parsing results and key parameters to generate operation instructions, and after verification, converts them into industrial protocol signals and sends them to the equipment controller.

[0173] In this embodiment, the intelligent decision-making and instruction generation unit 4 includes a model and rule configuration module 41, a job instruction generation module 42, an instruction verification module 43, and a protocol conversion and transmission module 44, wherein:

[0174] The model and rule configuration module 41 is used to store the trained gradient boosting tree model and configure the rigid rules related to dock operations; the configuration content of the model and rule configuration module 41 specifically includes:

[0175] The training data for the gradient boosting tree model comes from historical data on equipment operating thresholds, cargo space occupancy rates, and operation duration at the bulk cargo terminal.

[0176] The configuration rules of the job rule engine cover safe operation specifications, job priority rules, and equipment operation permission rules;

[0177] The collaborative logic between the gradient boosting tree model and the job rule engine is as follows: the gradient boosting tree model outputs dynamic job decision suggestions, the rule engine performs rigid constraint verification on the suggestions, and the gradient boosting tree model and the job rule engine work together to output decision results that meet the needs of the scenario, providing a basis for the generation of job instructions.

[0178] Specifically, the model and rule configuration module 41 undertakes the core functions of storing the gradient boosting tree model and configuring the rigid rules for the job, providing algorithmic support and compliance constraints for instruction generation. The specific operations are as follows:

[0179] The training data for the gradient boosting tree model comes from more than one year of real historical operation data of the bulk cargo terminal, covering equipment operation threshold data (crane rated lifting capacity, belt conveyor maximum speed, etc.), cargo space occupancy rate data (daily / time period cargo area idle ratio), and operation duration statistics (distribution of operation time for various scenarios).

[0180] The data preprocessing process is as follows:

[0181] First, outliers are eliminated using the "3σ criterion" ( The data was marked as abnormal and removed, among which For a single data entry, The mean of the data. (where the standard deviation is 0). Then, min-max standardization is used to map the data to the [0,1] interval. The standardization formula is:

[0182] ;

[0183] in, This represents the standardized data value, with a range of [0,1]. Represents the original data value; This represents the historical minimum value of the data in this dimension; This indicates the historical maximum value of the data in this dimension.

[0184] Then, the preprocessed data was divided into training and testing sets in a 7:3 ratio. The model was trained using a gradient boosting tree algorithm based on a regression task, and the squared loss function was selected.

[0185] ;

[0186] in For the true value, These are the model's predicted values. Hyperparameters were tuned using a grid search method, with the following search ranges: learning rate 0.01-0.2, number of decision trees 100-500, tree depth 3-10 layers, and minimum number of samples per leaf node 5-20. The optimal configuration was selected based on the parameter combination with the lowest mean squared error (MSE) on the test set. Five-fold cross-validation (dividing the training set into five equal parts, alternating between four parts for training and one part for validation, repeated five times, and averaging the MSE) ensured the model's generalization ability. The trained model was stored as a Pickle serialized file on an industrial-grade SSD, supporting version rollback and incremental training (new data was incorporated into the model quarterly).

[0187] Furthermore, the configuration rules of the job rule engine are represented using first-order predicate logic and are categorized into three types: safety operation specifications, job priority rules, and equipment operation permission rules. The rule base is stored in XML format. The rule engine supports dynamic addition, deletion, modification, and querying of rules, and uses a rule parser to convert natural language rules into machine-executable logical expressions.

[0188] Furthermore, the collaborative logic between the gradient boosting tree model and the job rule engine is as follows:

[0189] First, the gradient boosting tree model receives standardized input data and outputs dynamic job decision suggestions (such as equipment scheduling schemes). Recommended values ​​for operating parameters );

[0190] The rules engine will then substitute the suggestions into the rule base for Boolean validation. The validation formula is as follows:

[0191] ;

[0192] in For the verification results ( In accordance with the rules (Conflict exists) For the first The logical expression of the rule, The total number of rules;

[0193] Next, if Directly output the decision result; if Location conflict rules (such as) Safety threshold ), calculate conflict deviation :

[0194] ;

[0195] The formula for correcting model parameters is:

[0196] ;

[0197] in, This indicates the recommended values ​​for the adjusted operating parameters; As a correction factor, it is set at 0.8-1.0 according to the importance of the rules, and its value is based on the practical experience of front-line operations at bulk cargo terminals.

[0198] When the conflict rule is a safety-related rigid constraint (such as the crane's lifting capacity must not exceed the rated value, or the equipment's operating speed must not exceed the safety threshold), Set to 1.0 – Since safety is the bottom line for bulk cargo terminal operations, the parameters must be fully adjusted to the compliance range to avoid any safety risks;

[0199] When the conflict rule is an efficiency-related constraint (such as the priority of the work path or the recommended order of storage location allocation), Taking 0.8 – appropriately reserving 20% ​​adjustment space, which can meet the specifications and prevent deviation from the reasonable operating range due to excessive parameter adjustment, thus taking into account operating efficiency;

[0200] This value range is determined based on the core requirements of "safety first, efficiency appropriate" in the actual operation of bulk cargo terminals: if If the value is below 0.8, excessive parameter adjustments can easily affect the feasibility of the operation; if If the value is higher than 1.0, the violation parameter cannot be effectively corrected. Therefore, the range of 0.8-1.0 is selected to adapt to different types of rule constraints.

[0201] Adjust the suggested parameters using the model parameter correction formula above, and re-substitute them into the rule engine for verification until... Ultimately, collaborative decision-making results are output.

[0202] The job instruction generation module 42 receives the process analysis results output by the process node dynamic analysis unit 3, and generates preliminary job instructions by combining key job parameters and through the collaborative operation of the gradient boosting tree model and the job rule engine.

[0203] Specifically, the job instruction generation module 42 generates preliminary job instructions through a process of "data input - collaborative calculation - instruction formatting," and the specific operations are as follows:

[0204] The job instruction generation module 42 first receives the process parsing results (including current node ID, job priority, and resource constraint list) from the process node dynamic parsing unit 3, and simultaneously collects three types of key job parameters: real-time equipment operating status parameters (current equipment load, operating temperature, fault markers), cargo characteristic parameters (cargo weight, volume, hazardous materials identification), and resource allocation parameters (number of idle equipment, number of available personnel, and number of idle cargo positions). All input data undergoes secondary standardization according to model requirements (using the same min-max algorithm as model training), and weights are assigned according to feature importance to construct the model input vector:

[0205] ;

[0206] in, These are the feature weights, summing to 1, and are determined by the feature importance scores during model training.

[0207] Then, the gradient boosting tree model and the job rule engine are put into collaborative operation:

[0208] To start the gradient boosting tree model, input vector Input and output dynamic decision-making suggestions, including recommended equipment ID, job sequence, recommended operating parameter values, and estimated job duration. The Trec value is calculated by fusing model predictions with historical averages, using the following formula:

[0209] ;

[0210] in, This indicates the job duration predicted by the gradient boosting tree model; This indicates the historical average duration of similar tasks;

[0211] This represents the fusion coefficient, calibrated to 0.6 based on prediction accuracy (when the model prediction accuracy is higher than 80%).

[0212] The job rules engine then performs a comprehensive verification of the decision recommendations. The safety verification focuses on operational and environmental parameters (such as safety speed), the priority verification calculates the allocation weight using the "resource allocation priority coefficient and available resources," and the permission verification confirms the recommended devices. The degree of matching with the work team's permissions. If there is no conflict during the verification, the decision suggestion is directly used as the basic decision result; if there is a conflict, the work instruction generation module 42 triggers a secondary operation, adjusts the weights of the corresponding features in the input vector (such as increasing the weight of the safety parameter), and calls the model again to generate suggestions until all verifications are passed.

[0213] Finally, the operation instruction generation module 42 generates preliminary operation instructions according to the preset instruction format (which conforms to the communication protocol requirements of the terminal equipment). The instruction fields include instruction ID, operation object, operation task description, operation parameter details, execution time limit, and safety precautions. The instructions are temporarily stored in JSON format to ensure the integrity and readability of the fields.

[0214] The instruction verification module 43 performs safety threshold verification and permission verification on the preliminary operation instructions;

[0215] Specifically, the instruction verification module 43 is responsible for verifying the compliance of preliminary work instructions. Through dual logic of safety threshold verification and permission verification, it ensures that the instructions not only meet the safety baseline of bulk cargo terminal operations but also match the operating permissions of the work entity. The specific operation is as follows:

[0216] The instruction verification module 43 first performs a safety threshold verification on the preliminary operation instruction, using a method of "single parameter verification + multi-dimensional comprehensive verification":

[0217] Single-parameter verification calculates the parameter deviation rate for each operation parameter in the instruction. The formula is

[0218] ;

[0219] in Recommended values ​​for the parameters in the command. This refers to the safety threshold corresponding to the equipment or operation; the allowable range for this deviation rate is set at ≤5%, which is determined based on actual operational experience at bulk cargo terminals—the reasonable tolerance range for equipment operation is within this range, ensuring that misjudgments are not made due to minor fluctuations while also avoiding obvious risks of violations. Then the single-parameter validation passes. Marked as "Warning" It was judged as "not passed".

[0220] Subsequently, a multi-dimensional comprehensive verification is performed, and a security score is calculated through weighted summation:

[0221] ;

[0222] in The safety weights for the parameters are assigned (e.g., the weight for lifting capacity is 0.3, and the weight for wind speed is 0.25, based on the degree of influence of the parameters on operational safety). For single-parameter scores (pass = 100 points, warning = 60 points, fail = 0 points); when Time-sharing security check passed. The system outputs a warning and allows for manual confirmation; if the score is below 60, the verification fails.

[0223] After completing the security verification, the instruction verification module 43 performs the permission verification: based on the actual "work team-equipment type" permission management logic of the dock, a three-dimensional permission matrix is ​​constructed. ,in Indicates work group Equipment types Operation permissions Extract the team ID, equipment type, and operation type from the instruction, query the matrix element; if it is 1, the permission verification passes; otherwise, mark it as "permission mismatch" and determine it as failed. If both verifications pass, the instruction verification module 43 generates a verification pass mark and transmits the instruction; if either verification fails, the instruction is returned to the work instruction generation module 42, along with the reason for failure and adjustment suggestions, triggering the instruction secondary generation process.

[0224] The protocol conversion and transmission module 44 converts the verified operation instructions into industrial protocol signals and sends them to the dock operation equipment controller.

[0225] Specifically, the core responsibility of the protocol conversion and transmission module 44 is to convert the verified work instructions into industrial communication signals that can be recognized by the dock operation equipment, and to ensure reliable signal transmission and adapt to the communication characteristics of different equipment. The specific implementation is as follows:

[0226] The protocol conversion and transmission module 44 first identifies the communication protocol of the target device (the main industrial protocols used at the port are Modbus TCP and Profinet), and then executes the corresponding protocol conversion logic:

[0227] For the ModbusTCP protocol: The protocol conversion and transmission module 44 first converts the operation parameters in the instruction into 16-bit unsigned integers according to the device's preset scaling factor (for example, when the scaling factor is 10, 2m / s is converted to 20); then, according to the pre-stored "parameter-register address" mapping table, it assigns the corresponding register address to the converted parameters; finally, it constructs a ModbusTCP standard data frame, with the frame format being "7-byte MBAP header + Protocol Data Unit (PDU)", where the PDU contains the function code (0x06 for writing a single register, 0x10 for writing multiple registers), register address, data length, and converted parameter value.

[0228] For the Profinet protocol: the protocol conversion and transmission module 44 adopts the IO mapping method to associate the instruction parameters with the device's "Process Data Object (PDO)" and packs them into Ethernet frames according to the Profinet V2.3 specification to match the device's real-time communication requirements.

[0229] After the conversion is completed, the protocol conversion and sending module 44 performs CRC32 check on the data frame to ensure integrity: first, the CRC register is initialized to 0xFFFFFFFF, and each byte of the data frame is XORed with the CRC register in turn; then, the result of the operation is subjected to an 8-bit circular shift operation according to the CRC32 standard polynomial; finally, the CRC register after all bytes have been processed is inverted to obtain a 32-bit check code and appended to the end of the data frame.

[0230] Subsequently, the protocol conversion and transmission module 44 executes a differentiated transmission strategy based on the device type:

[0231] For equipment with high real-time requirements, such as cranes and stacker-reclaimers, the protocol conversion and transmission module 44 adopts a "real-time push mode": the transmission frequency is consistent with the equipment control cycle (50ms / time), and the timeout time is set to 1 second (to match the average response speed of the equipment).

[0232] For non-real-time equipment such as belt conveyors and weighbridges, the protocol conversion and sending module 44 adopts a "batch sending mode": the instructions are summarized and sent every 200ms, and the timeout is set to 3 seconds.

[0233] After transmission, the protocol conversion and transmission module 44 initiates a status monitoring mechanism: if an ACK receipt is received from the device controller within the timeout period, the transmission is considered successful; if no receipt is received, the protocol conversion and transmission module 44 automatically performs a retransmission operation (up to 3 retransmissions, with retransmission intervals of 500ms, 1000ms, and 1500ms respectively); if the transmission still fails after 3 retransmissions, the protocol conversion and transmission module 44 triggers a communication alarm in the bulk cargo terminal control system and backs up the unsuccessfully transmitted instructions (including data frames, checksums, and transmission logs) locally to a 64GB industrial-grade SD card (data is retained for 7 days). This facilitates technicians in troubleshooting equipment offline, network interruption, and other faults, ensuring that work instructions are accurately and timely transmitted to the target equipment.

[0234] The closed-loop control and verification unit 5 collects instruction execution feedback data, compares it with preset indicators, and generates a correction signal if the deviation exceeds the threshold. It also optimizes the collection frequency, confidence weight, node threshold and model parameters in reverse.

[0235] In this embodiment, the closed-loop control and verification unit 5 includes a feedback data acquisition module 51, an index comparison module 52, a correction signal generation module 53, and a parameter reverse optimization module 54, wherein:

[0236] The feedback data acquisition module 51 is used to collect instruction execution feedback data of the dock operation equipment, including operation completion rate, actual parameter execution value, and equipment operation status feedback data;

[0237] Specifically, the core function of the feedback data acquisition module 51 is to collect execution feedback data of bulk cargo terminal operation instructions from all dimensions, providing basic data support for subsequent indicator comparison and parameter optimization. The specific implementation is as follows: The feedback data acquisition module 51 collects three types of core data simultaneously by connecting to the status feedback interface of the bulk cargo terminal operation equipment controller and the progress statistics interface of the operation management system: First, operation completion data, covering the actual completion time of a single task, the percentage of cargo transfer / loading and unloading completed, and the actual flow time of nodes in the operation process; Second, actual parameter execution value data, corresponding to the operation parameters output by the intelligent decision-making and instruction generation unit 4, including the actual lifting weight of the crane, lifting speed, actual deviation of the running path, and actual conveying speed of the belt conveyor; Third, equipment operation status feedback data, including real-time equipment load rate, operating temperature, fault alarm codes, energy consumption data, etc.

[0238] Meanwhile, the feedback data acquisition module 51 adopts a "real-time + periodic" acquisition strategy: the equipment operation status feedback data is acquired in real time at 100ms / time (matching the equipment control cycle), and the work completion rate and actual parameter execution value data are acquired periodically at 5s / time (balancing data real-time performance and system overhead). After standardized encapsulation, the acquired data is temporarily stored in the local cache in JSON format to ensure the integrity and timeliness of the data.

[0239] The indicator comparison module 52 compares the feedback data with the preset operation indicators and calculates the data deviation value;

[0240] Specifically, the indicator comparison module 52 is responsible for quantitatively comparing the feedback data with preset work indicators to accurately identify deviations in the instruction execution process. The specific implementation is as follows:

[0241] After the feedback data acquisition module 51 transmits the collected feedback data to the indicator comparison module 52, the module first calls the pre-stored preset operation indicator library, where the preset indicators are configured differently according to the operation type (such as the preset completion time of loading and unloading operations is 30min / container area, the allowable range of crane lifting capacity execution error is ≤5%, and the safety threshold of equipment operating load rate is ≤80%).

[0242] Subsequently, the indicator comparison module 52 calculates the corresponding deviation value for different types of feedback data:

[0243] Deviation in assignment completion: ;in This refers to the actual completion time of a single task. The preset completion time for this task type;

[0244] Actual deviation of parameters: ;in These are the actual parameter values ​​executed by the device. Recommended values ​​for the parameters output by the intelligent decision-making and instruction generation unit 4;

[0245] Equipment condition deviation: ;in This refers to the real-time operating status values ​​of the equipment (such as load rate). This is the preset safety threshold corresponding to this state.

[0246] The index comparison module 52 compares the calculated deviation values ​​with the preset deviation thresholds, marks the deviation items that exceed the thresholds, and provides a basis for the subsequent generation of correction signals.

[0247] The correction signal generation module 53 is used to generate a targeted correction signal when the deviation value exceeds a set threshold.

[0248] Specifically, the correction signal generation module 53 generates a targeted correction signal based on the over-threshold deviation items marked by the index comparison module 52. The specific implementation is as follows: The correction signal generation module 53 first identifies the type and degree of influence of the deviation items, and generates a signal according to the mapping rule of "deviation type - correction direction":

[0249] like (Job timeout) Generate a correction signal to "shorten the execution time of the corresponding job instruction + optimize the equipment scheduling priority of similar jobs";

[0250] like (Parameter execution exceeds the limit), generate a correction signal to "narrow the recommended range of the corresponding operation parameter + strengthen the safety verification weight of the parameter";

[0251] like (Device status violation) Generate a correction signal to "reduce the upper limit of the load parameters of the corresponding device and increase the sampling frequency of the device's operating status".

[0252] The correction signal is encapsulated in the form of a structured instruction package, which includes three core fields: deviation type, correction target unit, and specific adjustment direction, to ensure the accuracy of subsequent parameter reverse optimization.

[0253] The parameter reverse optimization module 54 transmits the correction signal to the multi-source parameter acquisition unit 1, the multi-dimensional parameter adaptive fusion unit 2, the process node dynamic analysis unit 3, and the intelligent decision and instruction generation unit 4 respectively, and reverse optimizes the acquisition frequency of the multi-source parameter acquisition unit 1, the confidence weight of the multi-dimensional parameter adaptive fusion unit 2, the node threshold of the process node dynamic analysis unit 3, and the model parameters of the intelligent decision and instruction generation unit 4.

[0254] Specifically, the parameter reverse optimization module 54 is responsible for transmitting correction signals to each front-end unit to achieve dynamic iterative optimization of parameters throughout the entire process. The specific implementation is as follows: The parameter reverse optimization module 54 transmits correction signals according to the target unit, with the corresponding optimization logic being:

[0255] For multi-source parameter acquisition unit 1: the core optimization is "matching the frequency of deviations with the requirements for acquisition accuracy". For example, if the correction signal indicates "a certain device has violated the status three times in a row (continuous deviation)", and the operation type is hazardous chemical loading and unloading (high safety requirements), then the acquisition frequency of the core parameters of the device (such as load rate and operating temperature) is increased from 500ms / time to 100ms / time (significantly improving real-time performance); if the correction signal indicates "a certain device has only once exhibited redundant acquisition of non-core parameters (occasional deviation)", and the operation type is ordinary cargo handling (routine requirements), then the acquisition frequency of the non-core parameters (such as equipment shell temperature) is reduced from 500ms / time to 1s / time (reducing system overhead); if the correction signal indicates "the data acquisition delay of a certain area's cargo location causes node judgment deviation", then the acquisition frequency of the relevant parameters of the cargo location in that area is increased from 5s / time to 2s / time (balancing real-time performance and overhead).

[0256] For the multi-dimensional parameter adaptive fusion unit 2: the core optimization is "dynamically adjusting the confidence weights based on the reliability of the data source". For example, if the correction signal indicates that "a certain sensor (such as a weight sensor) has a deviation of more than 8% from the actual value for 5 consecutive feedback data (continuous low reliability)", then the confidence weight of that data source is reduced from 0.3 to 0.15, while the confidence weight of the same type of high-reliability sensor is increased from 0.3 to 0.45 (ensuring that the total weight is still 1); if the correction signal indicates that "a certain data source has only a single deviation (intermittent low reliability)", then its confidence weight is slightly adjusted from 0.3 to 0.25, without significantly adjusting the weights of other data sources; if the correction signal indicates that "the overall deviation of the fused data is small, but the reliability of the video surveillance data source decreases in a certain scenario (such as operation in rainy weather), then in this scenario, the confidence weight of the video surveillance data source is temporarily reduced from 0.2 to 0.1, while the weight of the LiDAR data source is simultaneously increased to 0.3.

[0257] For the dynamic analysis unit 3 of the process node: the core optimization is "adjusting the threshold amplitude according to the direction of the node judgment deviation". For example, if the correction signal points to "a certain operation node (such as 'manifest verification completed') being misjudged as completed twice in a row (actually not completed, missed judgment deviation)", and the operation type is loading and unloading (high process rigor requirement), then the baseline threshold of the node is increased from the original 0.8 to 0.88 to strengthen the judgment standard; if the correction signal points to "a certain node only has a single misjudgment (occasional deviation) and does not affect subsequent operations", then the baseline threshold is slightly adjusted to 0.82 to avoid over-adjustment leading to overly strict judgment; if the correction signal points to "a certain node has a large judgment deviation in a resource-scarce scenario (overall resource saturation > 0.8)", then in this scenario, the upper limit of the optimization coefficient is increased from 1.1 to 1.15 to further tighten the threshold.

[0258] For the intelligent decision-making and instruction generation unit 4, the core optimization is "adapting the model optimization method according to the type of deviation". For example, if the correction signal points to "the execution deviation of the crane lifting weight parameter exceeds 5% continuously (parameter decision deviation)", then the feedback data corresponding to this type of deviation (actual lifting weight, recommended instruction value, equipment load status) is used as incremental training data to supplement the training set of the gradient boosting tree model. At the same time, the model learning rate is reduced from 0.1 to 0.06 (improving model stability), and the number of decision trees is increased from 200 to 280 (improving model fitting accuracy). If the correction signal points to "large deviation in operation duration prediction (process decision deviation)", then "operation period (peak / off-peak)" and "freight yard congestion status" are added as model features, and the model is retrained. If the correction signal points to "large decision deviation only for a certain type of special goods (such as oversized goods)", then a sub-model is constructed separately for the feedback data of this type of goods to complement the original model.

[0259] After optimization, the parameter reverse optimization module 54 records the parameter items, adjustment range, corresponding correction signals and operation scenarios of this optimization, forming an optimization log (retained for 30 days). At the same time, after every 10 optimizations, the optimization effect of each parameter is automatically calculated (such as whether the deviation rate has decreased). If a certain type of optimization strategy fails to meet expectations for 3 consecutive times (the deviation rate does not decrease or even increases), the strategy adjustment is automatically triggered (such as changing the optimization direction or adjustment range) to ensure the dynamic iterative adaptability of the optimization logic.

[0260] Those skilled in the art will understand that the process of implementing all or part of the steps of the above embodiments can be carried out by hardware or by a program instructing the relevant hardware.

[0261] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A comprehensive control system for bulk cargo terminal operations based on multi-dimensional parameter sensing, characterized in that, include: Multi-source parameter acquisition unit (1) adopts a multi-source architecture that combines industrial bus, distributed sensing, machine vision, UWB positioning and NFC technology to collect equipment operating parameters, environmental and weighing data, cargo information, operator and vehicle trajectory, cargo handling and shift data, and output multi-source heterogeneous raw dataset. The multi-dimensional parameter adaptive fusion unit (2) receives the original dataset output by the multi-source parameter acquisition unit (1), and has an built-in adaptive confidence model based on Kalman filtering improvement. By configuring a constraint confidence time window for each type of parameter, it combines the time decay confidence weight and the working condition reliability factor to complete the data time dimension calibration. At the same time, it achieves noise filtering of multi-source heterogeneous data through physical constraint calibration, introduces electromagnetic environment dimension calibration, calculates the electromagnetic attenuation coefficient based on UWB signal strength, phase difference and multipath delay, dynamically adjusts the contribution of UWB data in the fusion weight, and finally completes the standardized fusion of multi-source data to output a high-confidence fusion sensing dataset. The multi-dimensional parameter adaptive fusion unit (2) includes a time dimension calibration module (21), an electromagnetic environment calibration module (23), and a standardized fusion output module (24), wherein: The time dimension calibration module (21) receives the multi-source heterogeneous raw dataset output by the multi-source parameter acquisition unit (1) and calls the adaptive confidence model to perform time dimension calibration; the calibration process of the time dimension calibration module (21) includes the following steps: S21.1 Based on the parameter types and operational accuracy requirements of multi-source heterogeneous raw datasets, the reliable time window of the constraint is calibrated through on-site testing at the dock, the window of the equipment operating parameters is adapted to the equipment response speed, and the window of the environment and weighing data is adapted to the data stability requirements. S21.2, Calculate the time decay confidence weights using the adaptive confidence model. , Data collection time difference Working condition adaptability coefficient Relatedness, fit coefficient Calibration was performed through actual dock operations; this involved using an adaptive confidence model, based on the data acquisition time difference. The relationship with the constrained credible time window is calculated using a piecewise nonlinear formula. The core formula is as follows: when hour: ； when hour: ； In the formula: The time decay confidence weight, with a value range of [0,1]; This refers to the data acquisition time difference, which is the difference between the data processing time and the acquisition time. To constrain the credible time window, i.e., the constrained credible time window in S21.1; The working condition adaptation coefficient is determined by actual measurement and calibration in the actual operation scenario of the bulk cargo terminal, and classified according to the core working conditions. It is a natural constant; S21.3, Apply time decay confidence weights Multiplying the corresponding parameter by the operating condition reliability factor yields the time dimension confidence weight of the parameter. The operating condition reliability factor is calibrated based on the statistical results of data credibility under different operating conditions at the dock. The electromagnetic environment calibration module (23) receives the output data from the UWB positioning and acquisition module (14) and performs electromagnetic dimension calibration of the UWB data; the calibration process of the electromagnetic environment calibration module (23) includes the following steps: S23.1 Receive the UWB signal strength output by the UWB positioning acquisition module (14) Phase difference and multipath delay Actual measurement data; S23.2 Calculate the electromagnetic attenuation coefficient based on the data collected in S23.1 , and , , This is related to the scene adaptation weight coefficient, which is calibrated by actual measurement according to the distribution of interference sources at the dock and classified by work area; among them, a weighted summation formula is used to calculate based on the received measured data and the scene adaptation weight coefficient. The core formula is as follows: ； In the formula: For UWB signal strength, Its standardized value is calculated as follows: ;in , The measured extreme values ​​of UWB signal strength in the general cargo terminal operation scenario; It is the measured UWB signal strength modulus value under the current bulk cargo terminal operation scenario; For the phase difference of the UWB signal, Its standardized value is calculated as follows: ; For UWB signal multipath delay, Its standardized value is calculated as follows: ;in This represents the measured maximum value of UWB multipath delay in a bulk cargo terminal operation scenario; To adapt the weight coefficients to the scenario and meet the requirements According to the distribution of interference sources at the bulk cargo terminal, the electromagnetic interference intensity and UWB data distortion of each area were collected, and the weight coefficients of the corresponding areas were determined by linear fitting. S23.3, Establish A correlation model with UWB data distortion is used to dynamically adjust the confidence weights of the time dimension of UWB data based on the correlation model. The contribution of UWB data fusion varies with... Adaptive adjustment to changes; The standardized fusion output module (24) connects with the processing results of the time dimension calibration module (21), the physical constraint noise reduction module (22), and the electromagnetic environment calibration module (23) to complete the standardized fusion of multi-source data and output a high-confidence fusion sensing dataset. The process of standardized fusion of multi-source data and output of a high-confidence fusion sensing dataset by the standardized fusion output module (24) includes the following steps: S24.1, parameters with time-dimension confidence weights output by the synchronous receiving time dimension calibration module (21), denoised data output by the physical constraint denoising module (22), and data output by the electromagnetic environment calibration module (23). Adjusted UWB data; S24.

2. Based on the time dimension confidence weights of each parameter, perform weighted fusion calculations on data of the same dimension. The fusion result is then compared with the parameter confidence weights. And directly related to the measured data; S24.

3. The min-max normalization method is used to process the fused data of all dimensions to form a standardized high-confidence fused perception dataset. The process node dynamic analysis unit (3) pre-stores the standard process flow chart, matches and fuses the perception data to determine the current operation node, and combines the resource status to optimize the node threshold, outputting the process analysis result with constraints; the process node dynamic analysis unit (3) includes a chart storage module (31) and a node matching module (32), wherein: The graph storage module (31) is used to pre-store the standard process graphs of operations classified by operation type. The operation types cover loading and unloading of ships, loading and unloading of trains, collection and distribution of automobiles at ports, handling and repackaging scenarios. The graphs contain key nodes and node feature parameters corresponding to each scenario. The node matching module (32) receives the high-confidence fusion perception dataset output by the multi-dimensional parameter adaptive fusion unit (2), extracts key parameters, and matches them with the node feature parameters of the standard workflow diagram to determine the current work node. The intelligent decision-making and instruction generation unit (4) combines the gradient boosting tree model and the operation rule engine, inputs the parsing results and key parameters to generate operation instructions, and after verification, converts them into industrial protocol signals and sends them to the equipment controller. The closed-loop control and verification unit (5) collects instruction execution feedback data, compares it with preset indicators, generates a correction signal if the deviation exceeds the threshold, and optimizes the collection frequency, confidence weight, node threshold and model parameters in reverse.

2. The bulk cargo terminal operation full-process control system based on multi-dimensional parameter perception as described in claim 1, characterized in that, The multi-source parameter acquisition unit (1) includes an industrial bus acquisition module (11), a distributed sensing module (12), a machine vision and NFC acquisition module (13), a UWB positioning acquisition module (14), and a sorting and shift data acquisition module (15), wherein: The industrial bus acquisition module (11) is based on the sensor interface of the dock operation equipment and adopts industrial bus communication technology to acquire the operating parameters of the equipment. The distributed sensing module (12) integrates environmental sensors and weighbridge equipment to collect environmental temperature and humidity, wind speed, visibility data and cargo weighing data; The machine vision and NFC acquisition module (13) acquires the work ticket number and work line status information based on machine vision recognition technology, and binds the cargo area, cargo location, stack number and cargo type and packaging type information with the help of NFC technology. The UWB positioning acquisition module (14) uses UWB positioning technology to track the real-time trajectory data of the operators and vehicles. The cargo handling and shift data acquisition module (15) synchronously acquires work shift information, collects core cargo handling data, and summarizes the collected data to form a multi-source heterogeneous original dataset.

3. The bulk cargo terminal operation full-process control system based on multi-dimensional parameter perception as described in claim 2, characterized in that, The multidimensional parameter adaptive fusion unit (2) also includes a physical constraint noise reduction module (22), which connects to the output data of the time dimension calibration module (21) to perform noise filtering of multi-source heterogeneous data.

4. The bulk cargo terminal operation full-process control system based on multi-dimensional parameter perception according to claim 3, characterized in that, The process node dynamic parsing unit (3) further includes a threshold optimization module (33) and a parsing result output module (34), wherein: The threshold optimization module (33) obtains the status data of the dock operation resources and optimizes the judgment threshold of the current operation node in combination with the status data of the dock operation resources. The parsing result output module (34) integrates the current job node information and the optimized threshold to output the process parsing result containing job priority and resource constraints.

5. The bulk cargo terminal operation full-process control system based on multi-dimensional parameter perception according to claim 4, characterized in that, The intelligent decision-making and instruction generation unit (4) includes a model and rule configuration module (41), a job instruction generation module (42), an instruction verification module (43), and a protocol conversion and transmission module (44), wherein: The model and rule configuration module (41) is used to store the trained gradient boosting tree model and configure the rigid rules related to dock operations; The job instruction generation module (42) receives the process analysis results output by the process node dynamic analysis unit (3), and generates preliminary job instructions by combining key job parameters and through the collaborative operation of the gradient boosting tree model and the job rule engine. The instruction verification module (43) performs security threshold verification and permission verification on the preliminary operation instruction; The protocol conversion and transmission module (44) converts the verified operation instructions into industrial protocol signals and sends them to the dock operation equipment controller.

6. The bulk cargo terminal operation full-process control system based on multi-dimensional parameter perception according to claim 5, characterized in that, The configuration content of the model and rule configuration module (41) specifically includes: The training data for the gradient boosting tree model comes from historical terminal operation data such as equipment operation thresholds, cargo space occupancy rates, and operation duration statistics. The configuration rules of the job rule engine cover safe operation specifications, job priority rules, and equipment operation permission rules; The collaborative logic between the gradient boosting tree model and the job rule engine is as follows: the gradient boosting tree model outputs dynamic job decision suggestions, the rule engine performs rigid constraint verification on the suggestions, and the gradient boosting tree model and the job rule engine work together to output decision results that meet the needs of the scenario, providing a basis for the generation of job instructions.

7. The bulk cargo terminal operation full-process control system based on multi-dimensional parameter perception according to claim 6, characterized in that, The closed-loop control and verification unit (5) includes a feedback data acquisition module (51), an index comparison module (52), a correction signal generation module (53), and a parameter reverse optimization module (54), wherein: The feedback data acquisition module (51) is used to collect instruction execution feedback data of the dock operation equipment, including operation completion rate, actual parameter execution value, and equipment operation status feedback data; The indicator comparison module (52) compares the feedback data with the preset operation indicators and calculates the data deviation value; The correction signal generation module (53) is used to generate a targeted correction signal when the deviation value exceeds a set threshold. The parameter reverse optimization module (54) transmits the correction signal to the multi-source parameter acquisition unit (1), the multi-dimensional parameter adaptive fusion unit (2), the process node dynamic analysis unit (3), and the intelligent decision and instruction generation unit (4), respectively, to reverse optimize the acquisition frequency of the multi-source parameter acquisition unit (1), the confidence weight of the multi-dimensional parameter adaptive fusion unit (2), the node threshold of the process node dynamic analysis unit (3), and the model parameters of the intelligent decision and instruction generation unit (4).

Images